Face Generation¶

In this project, you'll use generative adversarial networks to generate new images of faces.

Get the Data¶

You'll be using two datasets in this project:

MNIST
CelebA

Since the celebA dataset is complex and you're doing GANs in a project for the first time, we want you to test your neural network on MNIST before CelebA. Running the GANs on MNIST will allow you to see how well your model trains sooner.

If you're using FloydHub, set data_dir to "/input" and use the FloydHub data ID "R5KrjnANiKVhLWAkpXhNBe".

data_dir = './data'

# FloydHub - Use with data ID "R5KrjnANiKVhLWAkpXhNBe"
#data_dir = '/input'


"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper

helper.download_extract('mnist', data_dir)
helper.download_extract('celeba', data_dir)

Found mnist Data
Found celeba Data

Explore the Data¶

MNIST¶

As you're aware, the MNIST dataset contains images of handwritten digits. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
import os
from glob import glob
from matplotlib import pyplot

mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')

<matplotlib.image.AxesImage at 0x7f09c4d2e630>

CelebA¶

The CelebFaces Attributes Dataset (CelebA) dataset contains over 200,000 celebrity images with annotations. Since you're going to be generating faces, you won't need the annotations. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB')
pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB'))

<matplotlib.image.AxesImage at 0x7f09c4c126a0>

Preprocess the Data¶

Since the project's main focus is on building the GANs, we'll preprocess the data for you. The values of the MNIST and CelebA dataset will be in the range of -0.5 to 0.5 of 28x28 dimensional images. The CelebA images will be cropped to remove parts of the image that don't include a face, then resized down to 28x28.

The MNIST images are black and white images with a single color channel while the CelebA images have 3 color channels (RGB color channel).

Build the Neural Network¶

You'll build the components necessary to build a GANs by implementing the following functions below:

model_inputs
discriminator
generator
model_loss
model_opt
train

Check the Version of TensorFlow and Access to GPU¶

This will check to make sure you have the correct version of TensorFlow and access to a GPU

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf

# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer.  You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))

# Check for a GPU
if not tf.test.gpu_device_name():
    warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))

TensorFlow Version: 1.1.0
Default GPU Device: /gpu:0

Input¶

Implement the model_inputs function to create TF Placeholders for the Neural Network. It should create the following placeholders:

Real input images placeholder with rank 4 using image_width, image_height, and image_channels.
Z input placeholder with rank 2 using z_dim.
Learning rate placeholder with rank 0.

Return the placeholders in the following the tuple (tensor of real input images, tensor of z data)

import problem_unittests as tests

def model_inputs(image_width, image_height, image_channels, z_dim):
    """
    Create the model inputs
    :param image_width: The input image width
    :param image_height: The input image height
    :param image_channels: The number of image channels
    :param z_dim: The dimension of Z
    :return: Tuple of (tensor of real input images, tensor of z data, learning rate)
    """
    # TODO: Implement Function
    input_real = tf.placeholder(tf.float32, 
                                shape=(None, image_width, image_height, image_channels),
                                name='input_real')
    inputs_z = tf.placeholder(tf.float32, (None, z_dim), name='input_z')
    learning_rate = tf.placeholder(tf.float32, name='learning_rate')

    return input_real, inputs_z, learning_rate


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_inputs(model_inputs)

Tests Passed

Discriminator¶

Implement discriminator to create a discriminator neural network that discriminates on images. This function should be able to reuse the variables in the neural network. Use tf.variable_scope with a scope name of "discriminator" to allow the variables to be reused. The function should return a tuple of (tensor output of the discriminator, tensor logits of the discriminator).

alpha = 0.2
def leaky_relu(x):
    return tf.maximum(alpha*x,x)

def discriminator(images, reuse=False):
    """
    Create the discriminator network
    :param images: Tensor of input image(s)
    :param reuse: Boolean if the weights should be reused
    :return: Tuple of (tensor output of the discriminator, tensor logits of the discriminator)
    """
    # TODO: Implement Function
    with tf.variable_scope('discriminator', reuse=reuse):
        conv_layer1 = tf.layers.conv2d(images, 64, 5, strides=2, padding='same')
        relu1 = leaky_relu(conv_layer1)
        
        conv_layer2 = tf.layers.conv2d(conv_layer1, 128, 5, strides=2, padding='same')
        conv_layer2 = tf.layers.batch_normalization(conv_layer2, training=True)
        relu2 = leaky_relu(conv_layer2)
        
        conv_layer3 = tf.layers.conv2d(conv_layer2, 256, 5, strides=2, padding='same')
        conv_layer3 = tf.layers.batch_normalization(conv_layer3, training=True)
        relu3 = leaky_relu(conv_layer3)
        
        conv_layer4 = tf.layers.conv2d(conv_layer3, 512, 5, strides=2, padding='same')
        conv_layer4 = tf.layers.batch_normalization(conv_layer4)
        relu4 = leaky_relu(conv_layer4)
              
        #Flat
        flat = tf.reshape(relu4, (-1, 4*4*512))
        
        #Logits
        logits = tf.layers.dense(flat, 1)
        out = tf.sigmoid(logits)

    return out, logits


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_discriminator(discriminator, tf)

Tests Passed

Generator¶

Implement generator to generate an image using z. This function should be able to reuse the variables in the neural network. Use tf.variable_scope with a scope name of "generator" to allow the variables to be reused. The function should return the generated 28 x 28 x out_channel_dim images.

def generator(z, out_channel_dim, is_train=True):
    """
    Create the generator network
    :param z: Input z
    :param out_channel_dim: The number of channels in the output image
    :param is_train: Boolean if generator is being used for training
    :return: The tensor output of the generator
    """
    # TODO: Implement Function
    with tf.variable_scope('generator', reuse=False if is_train==True else True):
        #Fully connected layer
        fully_conn = tf.layers.dense(z, 7*7*512)
        fully_conn = tf.reshape(fully_conn, (-1,7,7,512))
        fully_conn = tf.layers.batch_normalization(fully_conn, training=is_train)
        fully_conn = leaky_relu(fully_conn)
        # 7x7x512

        #Convolutions
        conv_layer1 = tf.layers.conv2d_transpose(fully_conn, 256, 5, strides=2, padding='same')
        conv_layer1 = tf.layers.batch_normalization(conv_layer1, training=is_train)
        conv_layer1 = leaky_relu(conv_layer1)
        # 14x14x256

        conv_layer2 = tf.layers.conv2d_transpose(conv_layer1, 128, 5, strides=1, padding='same')
        conv_layer2 = tf.layers.batch_normalization(conv_layer2, training=is_train)
        conv_layer2 = leaky_relu(conv_layer2)
        # 14x14x128

        conv_layer3 = tf.layers.conv2d_transpose(conv_layer2, 64, 5, strides=2, padding='same')
        conv_layer3 = tf.layers.batch_normalization(conv_layer3, training=is_train)
        conv_layer3 = leaky_relu(conv_layer3)
        # 28x28x64 
        
        conv_layer4 = tf.layers.conv2d_transpose(conv_layer3, 32, 5, strides=1, padding='same')
        conv_layer4 = tf.layers.batch_normalization(conv_layer4, training=is_train)
        conv_layer4 = leaky_relu(conv_layer4)
        # 28x28x32 
        
        # Output layer
        logits = tf.layers.conv2d_transpose(conv_layer4, out_channel_dim, 5, strides=1, padding='same')
        # 28x28xout_dim 

        gen_out = tf.tanh(logits)

    return gen_out


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_generator(generator, tf)

Tests Passed

Loss¶

Implement model_loss to build the GANs for training and calculate the loss. The function should return a tuple of (discriminator loss, generator loss). Use the following functions you implemented:

discriminator(images, reuse=False)
generator(z, out_channel_dim, is_train=True)

def model_loss(input_real, input_z, out_channel_dim):
    """
    Get the loss for the discriminator and generator
    :param input_real: Images from the real dataset
    :param input_z: Z input
    :param out_channel_dim: The number of channels in the output image
    :return: A tuple of (discriminator loss, generator loss)
    """
    # TODO: Implement Function
    # Label smoothing 
    smooth = 0.1
    
    g_model = generator(input_z, out_channel_dim, is_train=True)
    d_model_real, d_logits_real = discriminator(input_real, reuse=False)
    d_model_fake, d_logits_fake = discriminator(g_model, reuse=True)
    
    d_loss_real = tf.reduce_mean(
        tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_real,
                                                labels=(tf.ones_like(d_model_real)*(1-smooth)) ))
    
    d_loss_fake = tf.reduce_mean(
        tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake,
                                                labels=tf.zeros_like(d_model_fake)))

    dicriminator_loss = d_loss_real + d_loss_fake
    
    g_loss = tf.reduce_mean(
        tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake, labels=tf.ones_like(d_model_fake)))

    
    return dicriminator_loss, g_loss


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_loss(model_loss)

Tests Passed

Optimization¶

Implement model_opt to create the optimization operations for the GANs. Use tf.trainable_variables to get all the trainable variables. Filter the variables with names that are in the discriminator and generator scope names. The function should return a tuple of (discriminator training operation, generator training operation).

def model_opt(d_loss, g_loss, learning_rate, beta1):
    """
    Get optimization operations
    :param d_loss: Discriminator loss Tensor
    :param g_loss: Generator loss Tensor
    :param learning_rate: Learning Rate Placeholder
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :return: A tuple of (discriminator training operation, generator training operation)
    """
    # TODO: Implement Function
    # Get weights and bias to update
    t_vars = tf.trainable_variables()
    d_vars = [var for var in t_vars if var.name.startswith('discriminator')]
    g_vars = [var for var in t_vars if var.name.startswith('generator')]

    # Optimize
    with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):
        d_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(d_loss, var_list=d_vars)
        g_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(g_loss, var_list=g_vars)
    
    return d_train_opt, g_train_opt


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_opt(model_opt, tf)

Tests Passed

Neural Network Training¶

Show Output¶

Use this function to show the current output of the generator during training. It will help you determine how well the GANs is training.

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np

def show_generator_output(sess, n_images, input_z, out_channel_dim, image_mode):
    """
    Show example output for the generator
    :param sess: TensorFlow session
    :param n_images: Number of Images to display
    :param input_z: Input Z Tensor
    :param out_channel_dim: The number of channels in the output image
    :param image_mode: The mode to use for images ("RGB" or "L")
    """
    cmap = None if image_mode == 'RGB' else 'gray'
    z_dim = input_z.get_shape().as_list()[-1]
    example_z = np.random.uniform(-1, 1, size=[n_images, z_dim])

    samples = sess.run(
        generator(input_z, out_channel_dim, False),
        feed_dict={input_z: example_z})

    images_grid = helper.images_square_grid(samples, image_mode)
    pyplot.imshow(images_grid, cmap=cmap)
    pyplot.show()

Train¶

Implement train to build and train the GANs. Use the following functions you implemented:

model_inputs(image_width, image_height, image_channels, z_dim)
model_loss(input_real, input_z, out_channel_dim)
model_opt(d_loss, g_loss, learning_rate, beta1)

Use the show_generator_output to show generator output while you train. Running show_generator_output for every batch will drastically increase training time and increase the size of the notebook. It's recommended to print the generator output every 100 batches.

def train(epoch_count, batch_size, z_dim, learning_rate, beta1, get_batches, data_shape, data_image_mode):
    """
    Train the GAN
    :param epoch_count: Number of epochs
    :param batch_size: Batch Size
    :param z_dim: Z dimension
    :param learning_rate: Learning Rate
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :param get_batches: Function to get batches
    :param data_shape: Shape of the data
    :param data_image_mode: The image mode to use for images ("RGB" or "L")
    """
    # TODO: Build Model
    print("Shape:", data_shape)
    print("Learning rate:", learning_rate)
    img_width = data_shape[1]
    img_height = data_shape[2]
    img_channels = data_shape[3]
    
    # Init GAN
    input_real, input_z, lr = model_inputs(img_width, img_height, img_channels, z_dim)
    print(input_real)
    print(input_z)
    d_loss, g_loss = model_loss(input_real, input_z, img_channels)
    d_opt, g_opt = model_opt(d_loss, g_loss, learning_rate, beta1)
    
    # Train
    steps = 0
    saver = tf.train.Saver()    
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        for epoch_i in range(epoch_count):
            for batch_images in get_batches(batch_size):
                # TODO: Train Model
                steps += 1
                
                #adjust the range
                batch_images *= 2
                
                batch_z = np.random.uniform(-1,1, size=(batch_size, z_dim))
                
                _ = sess.run(d_opt, feed_dict={input_real: batch_images, input_z: batch_z, lr: learning_rate})
                _ = sess.run(g_opt, feed_dict={input_real: batch_images, input_z: batch_z, lr: learning_rate})
                
                if steps % 10 == 0:
                    d_loss_train = d_loss.eval({input_z: batch_z, input_real: batch_images})
                    g_loss_train = g_loss.eval({input_z: batch_z})
                    
                    print("Epoch {}/{}...".format(epoch_i+1, epoch_count),
                         "Discriminator Loss: {:.4f}...".format(d_loss_train),
                         "Generator Loss: {:.4f}".format(g_loss_train))
                    
                if steps % 100 == 0:
                    show_generator_output(sess, 25, input_z, img_channels, data_image_mode)

MNIST¶

Test your GANs architecture on MNIST. After 2 epochs, the GANs should be able to generate images that look like handwritten digits. Make sure the loss of the generator is lower than the loss of the discriminator or close to 0.

batch_size = 64
z_dim = 100
learning_rate = 0.0002
beta1 = 0.5
alpha = 0.2

tf.reset_default_graph()
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 3

mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))

with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, mnist_dataset.get_batches,
          mnist_dataset.shape, mnist_dataset.image_mode)

Shape: (60000, 28, 28, 1)
Learning rate: 0.0002
Tensor("input_real:0", shape=(?, 28, 28, 1), dtype=float32)
Tensor("input_z:0", shape=(?, 100), dtype=float32)
Epoch 1/3... Discriminator Loss: 10.6759... Generator Loss: 0.0000
Epoch 1/3... Discriminator Loss: 0.3850... Generator Loss: 15.9446
Epoch 1/3... Discriminator Loss: 0.5147... Generator Loss: 25.3897
Epoch 1/3... Discriminator Loss: 0.4357... Generator Loss: 2.5811
Epoch 1/3... Discriminator Loss: 0.6435... Generator Loss: 29.5281
Epoch 1/3... Discriminator Loss: 0.3742... Generator Loss: 7.6315
Epoch 1/3... Discriminator Loss: 0.3522... Generator Loss: 17.8856
Epoch 1/3... Discriminator Loss: 0.5481... Generator Loss: 1.9654
Epoch 1/3... Discriminator Loss: 0.3720... Generator Loss: 3.4842
Epoch 1/3... Discriminator Loss: 0.7699... Generator Loss: 1.1998

Epoch 1/3... Discriminator Loss: 0.5224... Generator Loss: 1.8686
Epoch 1/3... Discriminator Loss: 0.3459... Generator Loss: 5.4047
Epoch 1/3... Discriminator Loss: 0.4485... Generator Loss: 2.6083
Epoch 1/3... Discriminator Loss: 0.4588... Generator Loss: 3.0327
Epoch 1/3... Discriminator Loss: 2.6748... Generator Loss: 10.9120
Epoch 1/3... Discriminator Loss: 0.5544... Generator Loss: 2.5647
Epoch 1/3... Discriminator Loss: 1.0995... Generator Loss: 0.8506
Epoch 1/3... Discriminator Loss: 0.7204... Generator Loss: 5.2743
Epoch 1/3... Discriminator Loss: 0.3776... Generator Loss: 3.9920
Epoch 1/3... Discriminator Loss: 0.9429... Generator Loss: 1.0773

Epoch 1/3... Discriminator Loss: 0.5948... Generator Loss: 2.3509
Epoch 1/3... Discriminator Loss: 0.7165... Generator Loss: 1.6678
Epoch 1/3... Discriminator Loss: 1.0319... Generator Loss: 0.8785
Epoch 1/3... Discriminator Loss: 1.2453... Generator Loss: 1.5639
Epoch 1/3... Discriminator Loss: 1.0666... Generator Loss: 0.8732
Epoch 1/3... Discriminator Loss: 1.1580... Generator Loss: 2.0450
Epoch 1/3... Discriminator Loss: 1.5884... Generator Loss: 0.4272
Epoch 1/3... Discriminator Loss: 1.2102... Generator Loss: 2.6429
Epoch 1/3... Discriminator Loss: 1.0838... Generator Loss: 1.0571
Epoch 1/3... Discriminator Loss: 1.1128... Generator Loss: 0.9062

Epoch 1/3... Discriminator Loss: 1.3905... Generator Loss: 0.5787
Epoch 1/3... Discriminator Loss: 1.4338... Generator Loss: 0.7125
Epoch 1/3... Discriminator Loss: 1.2327... Generator Loss: 0.9266
Epoch 1/3... Discriminator Loss: 1.4052... Generator Loss: 0.5118
Epoch 1/3... Discriminator Loss: 1.2747... Generator Loss: 0.6679
Epoch 1/3... Discriminator Loss: 1.3288... Generator Loss: 0.6086
Epoch 1/3... Discriminator Loss: 1.5997... Generator Loss: 0.4353
Epoch 1/3... Discriminator Loss: 1.5618... Generator Loss: 0.4598
Epoch 1/3... Discriminator Loss: 1.2131... Generator Loss: 0.7570
Epoch 1/3... Discriminator Loss: 1.2292... Generator Loss: 2.0198

Epoch 1/3... Discriminator Loss: 1.6250... Generator Loss: 1.4474
Epoch 1/3... Discriminator Loss: 1.1621... Generator Loss: 1.1622
Epoch 1/3... Discriminator Loss: 1.5011... Generator Loss: 1.2671
Epoch 1/3... Discriminator Loss: 1.1474... Generator Loss: 0.9683
Epoch 1/3... Discriminator Loss: 1.2908... Generator Loss: 0.8514
Epoch 1/3... Discriminator Loss: 1.2033... Generator Loss: 0.7248
Epoch 1/3... Discriminator Loss: 1.3642... Generator Loss: 1.0596
Epoch 1/3... Discriminator Loss: 1.0750... Generator Loss: 1.0363
Epoch 1/3... Discriminator Loss: 1.5917... Generator Loss: 0.4092
Epoch 1/3... Discriminator Loss: 1.3607... Generator Loss: 0.6106

Epoch 1/3... Discriminator Loss: 1.1410... Generator Loss: 1.3629
Epoch 1/3... Discriminator Loss: 1.0790... Generator Loss: 1.0599
Epoch 1/3... Discriminator Loss: 1.3014... Generator Loss: 1.3915
Epoch 1/3... Discriminator Loss: 1.1473... Generator Loss: 0.7982
Epoch 1/3... Discriminator Loss: 1.0507... Generator Loss: 0.9577
Epoch 1/3... Discriminator Loss: 1.1317... Generator Loss: 0.9290
Epoch 1/3... Discriminator Loss: 1.0548... Generator Loss: 0.9006
Epoch 1/3... Discriminator Loss: 1.2472... Generator Loss: 0.8714
Epoch 1/3... Discriminator Loss: 1.0646... Generator Loss: 1.1167
Epoch 1/3... Discriminator Loss: 1.0713... Generator Loss: 0.9914

Epoch 1/3... Discriminator Loss: 1.2303... Generator Loss: 0.8020
Epoch 1/3... Discriminator Loss: 1.2048... Generator Loss: 0.6414
Epoch 1/3... Discriminator Loss: 1.0745... Generator Loss: 0.8893
Epoch 1/3... Discriminator Loss: 1.2471... Generator Loss: 0.8446
Epoch 1/3... Discriminator Loss: 1.2796... Generator Loss: 0.6288
Epoch 1/3... Discriminator Loss: 1.2408... Generator Loss: 1.1503
Epoch 1/3... Discriminator Loss: 0.9170... Generator Loss: 1.1327
Epoch 1/3... Discriminator Loss: 0.9173... Generator Loss: 1.5018
Epoch 1/3... Discriminator Loss: 0.8387... Generator Loss: 1.4442
Epoch 1/3... Discriminator Loss: 0.9467... Generator Loss: 0.9882

Epoch 1/3... Discriminator Loss: 0.9649... Generator Loss: 1.1397
Epoch 1/3... Discriminator Loss: 0.9503... Generator Loss: 1.3441
Epoch 1/3... Discriminator Loss: 1.0965... Generator Loss: 0.9799
Epoch 1/3... Discriminator Loss: 1.1110... Generator Loss: 0.8335
Epoch 1/3... Discriminator Loss: 1.1207... Generator Loss: 0.8964
Epoch 1/3... Discriminator Loss: 1.1679... Generator Loss: 0.7057
Epoch 1/3... Discriminator Loss: 1.1336... Generator Loss: 1.1336
Epoch 1/3... Discriminator Loss: 0.9310... Generator Loss: 1.8620
Epoch 1/3... Discriminator Loss: 1.0531... Generator Loss: 1.2455
Epoch 1/3... Discriminator Loss: 1.0853... Generator Loss: 1.5548

Epoch 1/3... Discriminator Loss: 1.2976... Generator Loss: 0.5930
Epoch 1/3... Discriminator Loss: 1.3361... Generator Loss: 0.6169
Epoch 1/3... Discriminator Loss: 1.0241... Generator Loss: 1.0199
Epoch 1/3... Discriminator Loss: 1.5507... Generator Loss: 0.4776
Epoch 1/3... Discriminator Loss: 1.1874... Generator Loss: 0.6560
Epoch 1/3... Discriminator Loss: 1.1874... Generator Loss: 0.8100
Epoch 1/3... Discriminator Loss: 0.9670... Generator Loss: 1.0371
Epoch 1/3... Discriminator Loss: 1.3002... Generator Loss: 0.5240
Epoch 1/3... Discriminator Loss: 1.1526... Generator Loss: 1.0159
Epoch 1/3... Discriminator Loss: 1.0215... Generator Loss: 0.9527

Epoch 1/3... Discriminator Loss: 1.0544... Generator Loss: 1.7359
Epoch 1/3... Discriminator Loss: 1.1430... Generator Loss: 1.2899
Epoch 1/3... Discriminator Loss: 1.2080... Generator Loss: 1.1458
Epoch 2/3... Discriminator Loss: 1.0836... Generator Loss: 0.9371
Epoch 2/3... Discriminator Loss: 1.1306... Generator Loss: 0.8468
Epoch 2/3... Discriminator Loss: 1.2669... Generator Loss: 0.9564
Epoch 2/3... Discriminator Loss: 1.2021... Generator Loss: 1.1877
Epoch 2/3... Discriminator Loss: 1.2509... Generator Loss: 0.9788
Epoch 2/3... Discriminator Loss: 1.2513... Generator Loss: 0.7269
Epoch 2/3... Discriminator Loss: 1.1444... Generator Loss: 1.1469

CelebA¶

Run your GANs on CelebA. It will take around 20 minutes on the average GPU to run one epoch. You can run the whole epoch or stop when it starts to generate realistic faces.

batch_size = 64
z_dim = 100
learning_rate = 0.0002
beta1 = 0.2
alpha = 0.08

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 8

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)

Shape: (202599, 28, 28, 3)
Learning rate: 0.0002
Tensor("input_real:0", shape=(?, 28, 28, 3), dtype=float32)
Tensor("input_z:0", shape=(?, 100), dtype=float32)
Epoch 1/8... Discriminator Loss: 2.8972... Generator Loss: 0.0943
Epoch 1/8... Discriminator Loss: 0.3717... Generator Loss: 7.1293
Epoch 1/8... Discriminator Loss: 0.9751... Generator Loss: 0.8520
Epoch 1/8... Discriminator Loss: 0.5528... Generator Loss: 1.8244
Epoch 1/8... Discriminator Loss: 1.2481... Generator Loss: 0.6812
Epoch 1/8... Discriminator Loss: 1.9264... Generator Loss: 0.3464
Epoch 1/8... Discriminator Loss: 4.4561... Generator Loss: 0.0225
Epoch 1/8... Discriminator Loss: 1.4660... Generator Loss: 0.7730
Epoch 1/8... Discriminator Loss: 1.6436... Generator Loss: 0.8946
Epoch 1/8... Discriminator Loss: 1.1863... Generator Loss: 0.6426

Epoch 1/8... Discriminator Loss: 3.9317... Generator Loss: 0.0433
Epoch 1/8... Discriminator Loss: 1.6519... Generator Loss: 0.4812
Epoch 1/8... Discriminator Loss: 1.1996... Generator Loss: 0.6696
Epoch 1/8... Discriminator Loss: 0.8955... Generator Loss: 1.0986
Epoch 1/8... Discriminator Loss: 1.0411... Generator Loss: 2.0508
Epoch 1/8... Discriminator Loss: 0.8639... Generator Loss: 1.3387
Epoch 1/8... Discriminator Loss: 1.2124... Generator Loss: 1.0487
Epoch 1/8... Discriminator Loss: 1.7435... Generator Loss: 0.4459
Epoch 1/8... Discriminator Loss: 2.0444... Generator Loss: 0.2366
Epoch 1/8... Discriminator Loss: 2.0426... Generator Loss: 0.3098

Epoch 1/8... Discriminator Loss: 1.1957... Generator Loss: 0.6893
Epoch 1/8... Discriminator Loss: 0.8082... Generator Loss: 1.1730
Epoch 1/8... Discriminator Loss: 1.1649... Generator Loss: 0.7060
Epoch 1/8... Discriminator Loss: 1.0622... Generator Loss: 0.9095
Epoch 1/8... Discriminator Loss: 0.7906... Generator Loss: 1.2013
Epoch 1/8... Discriminator Loss: 1.7398... Generator Loss: 0.3446
Epoch 1/8... Discriminator Loss: 0.6544... Generator Loss: 1.4862
Epoch 1/8... Discriminator Loss: 0.4062... Generator Loss: 3.7250
Epoch 1/8... Discriminator Loss: 1.2355... Generator Loss: 2.3402
Epoch 1/8... Discriminator Loss: 0.5916... Generator Loss: 2.0761

Epoch 1/8... Discriminator Loss: 1.4884... Generator Loss: 0.4464
Epoch 1/8... Discriminator Loss: 0.8268... Generator Loss: 1.6319
Epoch 1/8... Discriminator Loss: 0.7717... Generator Loss: 1.1933
Epoch 1/8... Discriminator Loss: 0.9675... Generator Loss: 0.9621
Epoch 1/8... Discriminator Loss: 0.9946... Generator Loss: 2.5354
Epoch 1/8... Discriminator Loss: 1.3584... Generator Loss: 0.5345
Epoch 1/8... Discriminator Loss: 0.9798... Generator Loss: 1.0858
Epoch 1/8... Discriminator Loss: 1.6005... Generator Loss: 0.4516
Epoch 1/8... Discriminator Loss: 0.6353... Generator Loss: 2.8181
Epoch 1/8... Discriminator Loss: 0.7098... Generator Loss: 1.4147

Epoch 1/8... Discriminator Loss: 1.3756... Generator Loss: 0.5569
Epoch 1/8... Discriminator Loss: 0.4414... Generator Loss: 3.5374
Epoch 1/8... Discriminator Loss: 0.7417... Generator Loss: 1.4692
Epoch 1/8... Discriminator Loss: 1.7294... Generator Loss: 0.4187
Epoch 1/8... Discriminator Loss: 2.7799... Generator Loss: 0.1166
Epoch 1/8... Discriminator Loss: 1.3268... Generator Loss: 0.5031
Epoch 1/8... Discriminator Loss: 1.0476... Generator Loss: 2.1691
Epoch 1/8... Discriminator Loss: 1.1119... Generator Loss: 0.7700
Epoch 1/8... Discriminator Loss: 1.3601... Generator Loss: 0.5366
Epoch 1/8... Discriminator Loss: 2.4618... Generator Loss: 0.1544

Epoch 1/8... Discriminator Loss: 2.2731... Generator Loss: 0.1977
Epoch 1/8... Discriminator Loss: 2.1222... Generator Loss: 0.2313
Epoch 1/8... Discriminator Loss: 1.0150... Generator Loss: 0.8297
Epoch 1/8... Discriminator Loss: 1.2528... Generator Loss: 0.6126
Epoch 1/8... Discriminator Loss: 1.4400... Generator Loss: 0.4595
Epoch 1/8... Discriminator Loss: 1.0566... Generator Loss: 1.0718
Epoch 1/8... Discriminator Loss: 1.1569... Generator Loss: 0.7016
Epoch 1/8... Discriminator Loss: 1.4454... Generator Loss: 2.5242
Epoch 1/8... Discriminator Loss: 2.2191... Generator Loss: 0.1853
Epoch 1/8... Discriminator Loss: 1.6707... Generator Loss: 0.3414

Epoch 1/8... Discriminator Loss: 1.0159... Generator Loss: 0.9349
Epoch 1/8... Discriminator Loss: 0.8971... Generator Loss: 1.5615
Epoch 1/8... Discriminator Loss: 0.8695... Generator Loss: 1.0526
Epoch 1/8... Discriminator Loss: 1.9380... Generator Loss: 0.2572
Epoch 1/8... Discriminator Loss: 1.1264... Generator Loss: 2.8170
Epoch 1/8... Discriminator Loss: 1.0362... Generator Loss: 0.8936
Epoch 1/8... Discriminator Loss: 1.2471... Generator Loss: 0.5932
Epoch 1/8... Discriminator Loss: 0.9678... Generator Loss: 0.8840
Epoch 1/8... Discriminator Loss: 1.5487... Generator Loss: 0.3674
Epoch 1/8... Discriminator Loss: 2.0427... Generator Loss: 0.2580

Epoch 1/8... Discriminator Loss: 0.6702... Generator Loss: 2.1132
Epoch 1/8... Discriminator Loss: 1.3189... Generator Loss: 0.4913
Epoch 1/8... Discriminator Loss: 1.1075... Generator Loss: 0.7246
Epoch 1/8... Discriminator Loss: 0.8829... Generator Loss: 1.0012
Epoch 1/8... Discriminator Loss: 0.9890... Generator Loss: 0.9639
Epoch 1/8... Discriminator Loss: 1.7359... Generator Loss: 0.3087
Epoch 1/8... Discriminator Loss: 1.0972... Generator Loss: 0.6916
Epoch 1/8... Discriminator Loss: 0.6967... Generator Loss: 1.7754
Epoch 1/8... Discriminator Loss: 0.8191... Generator Loss: 2.0154
Epoch 1/8... Discriminator Loss: 1.5255... Generator Loss: 2.6242

Epoch 1/8... Discriminator Loss: 1.4262... Generator Loss: 0.4607
Epoch 1/8... Discriminator Loss: 1.6904... Generator Loss: 0.3223
Epoch 1/8... Discriminator Loss: 1.3255... Generator Loss: 0.5162
Epoch 1/8... Discriminator Loss: 1.1388... Generator Loss: 0.6553
Epoch 1/8... Discriminator Loss: 1.5674... Generator Loss: 0.3732
Epoch 1/8... Discriminator Loss: 1.7581... Generator Loss: 0.3396
Epoch 1/8... Discriminator Loss: 0.9733... Generator Loss: 0.9915
Epoch 1/8... Discriminator Loss: 1.1078... Generator Loss: 2.3570
Epoch 1/8... Discriminator Loss: 0.9254... Generator Loss: 1.4141
Epoch 1/8... Discriminator Loss: 1.4124... Generator Loss: 0.4416

Epoch 1/8... Discriminator Loss: 1.6507... Generator Loss: 2.7155
Epoch 1/8... Discriminator Loss: 1.6887... Generator Loss: 0.3444
Epoch 1/8... Discriminator Loss: 1.0890... Generator Loss: 0.6891
Epoch 1/8... Discriminator Loss: 0.5776... Generator Loss: 2.0815
Epoch 1/8... Discriminator Loss: 1.4856... Generator Loss: 0.4342
Epoch 1/8... Discriminator Loss: 0.6826... Generator Loss: 2.9883
Epoch 1/8... Discriminator Loss: 1.9443... Generator Loss: 0.2523
Epoch 1/8... Discriminator Loss: 1.3499... Generator Loss: 0.4878
Epoch 1/8... Discriminator Loss: 0.8269... Generator Loss: 1.0786
Epoch 1/8... Discriminator Loss: 0.9972... Generator Loss: 0.9621

Submitting This Project¶

When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_face_generation.ipynb" and save it as a HTML file under "File" -> "Download as". Include the "helper.py" and "problem_unittests.py" files in your submission.